Skip to main content

Agent Actions

An agentic workflow engine that runs in your terminal. Define LLM pipelines in YAML, and the engine handles orchestration, validation, and error recovery.

Get started in 30 seconds​

Prerequisites: Python 3.11+ and an API key from any supported provider (OpenAI, Anthropic, Gemini, Groq, Mistral, Ollama).

Install Agent Actions:

uv pip install agent-actions

# Or with pip
pip install agent-actions

Start using Agent Actions:

cd your-project
agac run -a my_workflow

That's it! Continue with Quickstart (5 minutes) →

See Installation for configuration options or Troubleshooting if you hit issues.

What to build​

Incident triage — Classify severity, assess impact, assign teams, generate response plans. Parallel evaluators with consensus aggregation.

Contract review — Split contracts into clauses, analyze each for risk in parallel, aggregate findings into a unified report, and produce an executive summary.

Review analysis — Multi-model pipeline for product reviews — extract claims, score quality via parallel consensus, draft merchant responses, and surface product insights.

Catalog enrichment — Take raw book/product data, enrich with classifications, marketing copy, SEO keywords, and recommendations.

What Agent Actions does for you​

Build pipelines from YAML: Define your workflow in plain YAML. Agent Actions handles DAG orchestration, parallelization, and dependency resolution.

Validate every output: Every LLM response is validated against JSON Schema. Invalid outputs trigger automatic reprompting until they conform.

Mix and match providers: Chain OpenAI, Anthropic, Gemini, Groq, Mistral, and Ollama in the same workflow. Switch models per-action.

Catch errors before they cost you: Pre-flight validation checks your config, variables, and dependency wiring before any API calls are made.

How it works​

Write a YAML config:

actions:
  - name: extract
    prompt: "Extract key facts from: {{ source.content }}"
    schema: facts_schema

  - name: summarize
    dependencies: extract
    prompt: "Summarize: {{ extract.facts }}"

Run it:

agac run -a my_workflow

Next steps​